Runnable scripts that exercise the public API. Each writes a .wav (ignored by
git) to the current directory.
| Script | What it shows |
|---|---|
01_quickstart.py |
The minimal text → audio call. |
02_sound_effects.py |
Batch generation with classifier-free guidance. |
03_custom_schedule.py |
Building a pipeline with a cosine schedule + v-prediction. |
04_mel_roundtrip.py |
Mel analysis/synthesis via Griffin-Lim (no diffusion). |
05_interpolation.py |
Spherical latent interpolation (slerp) between two latents. |
Run any of them from the repo root:
python examples/01_quickstart.pyThe shipped weights are random, so generated audio is noise-shaped until the model is trained. The round-trip example (
04) produces real, recognisable audio because it does not involve the diffusion model.